YOLOv5-Based Object Detection for Emergency Response in Aerial Imagery

TL;DR

The paper tackles emergency-response object detection in aerial imagery by adapting YOLOv5 to detect critical objects such as ambulances, police vehicles, and fire engines from drone data. It presents a custom dataset with careful annotation, tailored architectural tweaks (anchor adjustments and higher input resolution), and a comprehensive evaluation showing strong real-time performance yet class-specific limitations for small or rare objects. The study demonstrates that YOLOv5 offers an attractive balance of speed and accuracy for real-time emergency response, with detailed insights into data needs, model improvements, and practical deployment considerations. These findings support deploying aerial-object detectors on edge devices for disaster management, traffic control, and urban planning, while outlining avenues for temporal analysis and architectural enhancements to further boost performance.

Abstract

This paper presents a robust approach for object detection in aerial imagery using the YOLOv5 model. We focus on identifying critical objects such as ambulances, car crashes, police vehicles, tow trucks, fire engines, overturned cars, and vehicles on fire. By leveraging a custom dataset, we outline the complete pipeline from data collection and annotation to model training and evaluation. Our results demonstrate that YOLOv5 effectively balances speed and accuracy, making it suitable for real-time emergency response applications. This work addresses key challenges in aerial imagery, including small object detection and complex backgrounds, and provides insights for future research in automated emergency response systems.

Figures (8)

• Figure 1: Training and validation metrics of the YOLOv5 model over 100 epochs.
• Figure 2: The model detects a flipped car with 0.50 confidence, highlighting challenges in distinguishing rare objects due to background complexity and irregular shapes.
• Figure 3: The model detects a fire engine (0.86 confidence) and a car on fire (0.52 confidence), showcasing strong performance for distinct objects but struggles with smaller, visually complex ones.
• Figure 4: The model detects two ambulances (0.51, 0.44) and a fire engine (0.38), highlighting challenges with occlusion and complex backgrounds.
• Figure 5: The model detects two fire engines with confidence scores of 0.88 and 0.78, demonstrating strong performance for large, distinct objects.